I build systems that outlast the person operating them — AI platforms running in production, and the institutions that keep them running. Most of it is built where infrastructure is unreliable and budgets are real. The standard is still global.
I build solutions that work and lead teams that deliver. My focus is on technology, social impact, and human development, spanning digital products used by thousands, non-profit work touching over 7,000 lives, and leadership programmes training the next generation.
Proficiency matters. I focus on what I can build, who I can serve, and what systems I can put in place to create value that outlasts me.
"Principles are the same, methods can be different."
Read Full Profile
Four things I've come to hold, from building in places where the infrastructure doesn't cooperate.
Knowledge is global. Whatever can be measured is global. There is a standard to doing anything well, and the information is available to whoever is willing to consume it. Where you build from is not an excuse for what you build.
Most decisions get made against the next paycheck, and that horizon produces work that has to be redone. I would rather spend five years building structure that runs than spend a career restarting.
If a process needs me present to work, it is not a system — it is a habit. Enforce the rules programmatically. Measure outcomes, not attendance. Make progress visible without anyone having to ask.
The distance between competent and authoritative is almost always time nobody billed for — reading the book, running the experiment, shipping the thing nobody asked for. Tools have made that mile shorter. Very few people still walk it.
AI platforms operating under real latency budgets, provider quotas, and cost ceilings — not demos.
Financial intelligence over a 14,000-record knowledge graph. Retrieval grounded in PostgreSQL; batch extraction parallelised from ~22s to ~7–8s to hold a 12-second pipeline budget.
Admissions scoring that reads readiness, not polish — four diagnostic probes through an LLM, with hard rules on what the model is never allowed to decide.
Inference infrastructure — a self-hosted gateway over Ollama on GPU, a hand-built batch client for ~50% lower cost, and vision extraction where text parsing gives up.
Multi-tenant content pipeline. Deterministic scoring filters candidates before any model spend; a style firewall gates output with automated edit retries.
Structures built to keep operating after the founder's attention moves elsewhere.
Technology company. Digital products across education, certification, fintech, and content.
diranx.comEmpowering 2,000+ widows and 5,000+ orphans in Cameroon over 7 years.
sustainingpeacefoundation.orgBreaking poverty cycles through digital skills, survival support, and mentorship since 2016.
cycle28.orgBuilt and maintained independently. Several are free and run without commercial intent.
Technical briefs and field notes from systems in production — written for the people who have to act on them.
I advise teams putting AI and software into production — particularly where bandwidth, budget, and infrastructure are real constraints rather than footnotes. I also speak and write on building systems that run without supervision.